Honestly, the biggest improvements left in CPU performance comes from massively increasing the caches. I don't want more cores. I want a freaking massive L3 cac

Honestly, the biggest improvements left in CPU performance comes from massively increasing the caches. I don't want more cores. I want a freaking massive L3 cache. I want direct access to insanely fast SSD's. For anything that isn't purely focused on floating point performance (where throwing a bunch of GPUs at the problem is fine) memory and storage are now bottlenecks that can still give us several order of magnitude improvements in performance. Then there's the smart move of integrating FGPAs on-die and connecting them with insanely fast buses. Now we can do cool things like process data coming in via a 40GBe link on a single on-die FPGA and perform business logic on the CPU with data that the FPGA writes to memory. OR write an FPGA accelerator for a function. Single core performance may not get better... but we still have huge places to make improvements that will require some changes in how we run code.